struct/union/pointer query (Why isn't this legal?)

[Chris Torek]

In article <1298 at motmpl.UUCP> ron at motmpl.UUCP (Ron Widell) writes:
>Please respond via e-mail, I will summarize.

Good suggestion, but I will ignore it.  Anyway:

>... As I interpret K&R, 1e (page 120 in particular); I think [the
>following code, with #ifdef's edited away] should be legal, so why
>won't any of the compilers I've tried accept it?

>int main() {
>	union {
>		unsigned short foo;
>		struct {
>			unsigned char fuz;
>			unsigned char bar;
>			} baz, *sptr;
>	} glorp, *uptr;
>
>	uptr = &glorp;
>	sptr = &glorp.baz;
>	uptr->foo = (unsigned short)0x4142;

The best way to understand what has been declared above is to `unwind'
it.  This requires adding tags to any structure or union declarations
that do not already have them.  Here, the inner `struct' (containing
`fuz' and `bar') has no tag, so add one.  At the same time, move the
declaration out:

	struct ss {
		unsigned char fuz;
		unsigned char bar;
	};

Likewise, the union has no tag, so add one:

	union uu {
		unsigned short foo;
		struct ss baz, *sptr;
	};

Next, go through all the struct and union declarations and expand
type names for all members.  Here only the union has any commas:

	union uu {
		unsigned short foo;
		struct ss baz;
		struct ss *sptr;
	};

Now that all structures and unions are fully named, add the
main() function and its local variables.  What we now have is this:

	struct ss {
		unsigned char fuz;
		unsigned char bar;
	};

	union uu {
		unsigned short foo;
		struct ss baz;
		struct ss *sptr;
	};

	main()
	{
		union uu glorp, *uptr;

		uptr = &glorp;
-->		sptr = &glorp.baz;
		uptr->foo = (unsigned short)0x4142;

		[the rest is missing, unnecessary]

The line marked with the arrow (-->) is where things go wrong.  It
should be obvious at this point that main() has only two local
variables, one (`glorp') of type `union uu' and one (`uptr') of type
`pointer to union uu'.  The local variable `sptr' simply does not
exist.  (glorp contains a field called sptr, however.)

>... I'm only interested in the legality of the union{..struct{..};};
>definition/declaration.  What I *think* I'm defining is a union with
>three members: 1) an unsigned short ... . 2) a structure of 2 unsigned
>chars. 3) a pointer to a structure of 2 unsigned chars.

That is indeed what you defined.
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163)
Domain:	chris at mimsy.umd.edu	Path:	uunet!mimsy!chris